Initialising Workbench...
Black out emails, tokens, and personal details in a screenshot before you share it. The image is never uploaded — detection and export both run in this tab.
Drop a file here or click to upload
PNG, JPG, WebP, BMP or GIF — or paste a screenshot with Ctrl/⌘ + V
Boxes are burned into the pixels of a freshly rendered PNG, so the exported file has no hidden layer to peel back and no original underneath. Re-encoding also drops the original EXIF block — GPS coordinates, device model, and capture time do not survive the export.
Text detection and the redacted export both run in this tab — the image is read from local memory and no copy is ever uploaded.
The Image Redactor covers up the parts of a screenshot you did not mean to share. It reads the image with a WebAssembly build of Tesseract OCR, turns every recognised word into a clickable box, and burns your selections into a freshly rendered PNG — all without the file leaving your browser.
Paste a screenshot straight from the clipboard and the scan starts on its own. From there, either click words individually or let auto-detect select the ones that look like emails, API keys, bearer tokens, JWTs, card numbers, or IP addresses — the same pattern set behind the Obfuscate & Redact tool. Anything OCR cannot read, including faces, avatars, QR codes, and logos, can be covered with a box you drag yourself.
Choose a solid block when the content is genuinely sensitive: the covered pixels are overwritten before the PNG is encoded, so there is no layer to peel back. Pixelate and blur are offered for softer edits, but both only transform the original pixels and have been reversed before, so they are the wrong choice for a password or a key. The export is re-encoded from a canvas, which also drops the original EXIF metadata — GPS location, device model, and capture time included.
No. The image is read from local memory, text detection runs as a WebAssembly build of Tesseract inside a Web Worker, and the redacted PNG is rendered on a canvas in the same tab. The engine and language models are served from this site rather than a third-party CDN, so nothing — not even the fact that you opened an image — leaves your device.
Not from a solid-block export. The image is redrawn onto a fresh canvas and the covered pixels are overwritten before the PNG is encoded, so there is no layer to hide, no crop to undo, and no original copy inside the file. Pixelate and blur are different — see below.
No. Both transform the original pixels rather than replacing them, and researchers have repeatedly recovered text from pixelated and blurred images — short strings in a known font, like a name or a six-digit code, are the easiest to reverse. Use them for aesthetics and use a solid block for anything you would not want read.
It runs the recognised text through the same regular-expression pattern set as the Obfuscate & Redact tool — emails, phone numbers, API keys, bearer tokens, JWTs, private-key headers, credit card numbers, IP addresses, and similar shapes — and selects the word boxes that overlap a match. It is a heuristic first pass, not a guarantee.
Auto-detect can only see what OCR read. Handwriting, stylised logos, low-contrast or very small UI text, and rotated text are often not recognised at all, and multi-line values such as a wrapped private key or a postal address are not matched because detection runs per visual line. Anything OCR missed can still be covered with a hand-drawn box.
Yes. Drag anywhere on the preview to draw a box over a face, an avatar, a QR code, a logo, or a chart. Hand-drawn boxes are exported exactly like word boxes, and hovering one shows a control to remove it.
Yes, as a side effect. The exported PNG is encoded fresh from a canvas, so the original EXIF block does not carry over — GPS coordinates, camera or phone model, and capture timestamps are all dropped along with any other embedded metadata.
PNG, JPG, WebP, BMP, and GIF go in; the export is always PNG. PNG is lossless, so the un-redacted parts of the image stay pixel-identical to the original and text elsewhere in the screenshot does not pick up compression artefacts.